Human Reproduction, Vol.30, No.5 pp. 1246 –1255, 2015 Advanced Access publication on March 18, 2015 doi:10.1093/humrep/dev063
ORIGINAL ARTICLE Reproductive epidemiology
Caffeine and caffeinated beverage consumption and risk of spontaneous abortion 1
Department of Epidemiology, Boston University School of Public Health, Boston, MA 02118, USA 2Slone Epidemiology Center, Boston University, Boston, MA 02215, USA 3RTI Health Solutions, Research Triangle Park, NC 12194 USA 4Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark *Correspondence address. E-mail: [email protected]
Submitted on December 9, 2014; resubmitted on February 18, 2015; accepted on February 26, 2015
study question: Is caffeine and caffeinated beverage consumption associated with the risk of spontaneous abortion (SAB)? summary answer: While preconceptional caffeine consumption was not materially associated with an increased risk of SAB, consumption during early pregnancy was associated with a small increased risk of SAB, although the relation was not linear.
what is known already: Caffeine has been hypothesized as a risk factor for SAB since the 1980s; however, results from previous studies have been conflicting.
study design, size, duration: This prospective cohort study included 5132 Danish women planning pregnancy and enrolled from 2007 to 2010.
participants/materials, setting, methods: Participants were women who conceived after entry into the Snart-Gravid cohort and who were aged 18 –40, in a stable relationship with a male partner, and did not use fertility treatments to conceive. Women reported their daily caffeine and caffeinated beverage consumption on questionnaires before conception and during early pregnancy. All exposure measurements were prospective with respect to outcome ascertainment. We estimated hazard ratios (HRs) of SAB for categories of caffeine consumption in milligrams (mg) per day and the corresponding 95% confidence intervals (CIs) using Cox proportional hazards regression models with gestational weeks as the time scale. main results and the role of chance: There were 732 women (14.3%) who were identified as having a SAB. In the preconceptional period, caffeine consumption was not materially associated with SAB risk (HR comparing ≥300 with ,100 mg/day: 1.09; 95% CI: 0.89, 1.33). In early pregnancy, the HRs for 100 –199, 200 –299 and ≥300 mg/day of caffeine consumption were 1.62 (95% CI: 1.19, 2.22), 1.48 (95% CI: 1.03, 2.13) and 1.23 (95% CI: 0.61, 2.46), respectively, compared with that for ,100 mg/day. limitations, reasons for caution: The observed results may be affected by non-differential exposure misclassification, reverse causation and residual confounding.
wider implications of the findings: This is the largest study to date of prospectively measured, preconception caffeine consumption and risk of SAB. We were able to reduce the likelihood of differential left truncation bias and recall bias present in other analyses.
study funding/competing interest(s): Snart-Gravid was funded by the NICHD (R21-050264). Dr. Hahn’s work was funded in part by the BU Reproductive, Perinatal, and Pediatric Epidemiology Training Grant NIH #T32HD052458. There are no competing interests. Key words: caffeine / coffee / spontaneous abortion / cohort study
Introduction Caffeine crosses the placenta (Dlugosz and Bracken, 1992) and has a prolonged metabolism in pregnant women (15.08 hour half-life) compared
with non-pregnant women (4.71 hour half-life). Fetuses eliminate caffeine very slowly, suggesting that maternal caffeine ingestion could increase fetal caffeine levels exponentially (Brazier et al., 1983). Further, some studies (Lucero et al., 2001; Lawson et al., 2002; Kotsopoulos
& The Author 2015. Published by Oxford University Press on behalf of the European Society of Human Reproduction and Embryology. All rights reserved. For Permissions, please email: [email protected]
Downloaded from http://humrep.oxfordjournals.org/ at University of California, San Francisco on April 17, 2015
K.A. Hahn 1,*, L.A. Wise 1,2, K.J. Rothman 1,3, E.M. Mikkelsen4, S.B. Brogly 1, H.T. Sørensen 1,4, A.H. Riis 4, and E.E. Hatch 1
1247
Caffeine consumption and spontaneous abortion
Materials and Methods Data collection The Snart-Gravid study is an Internet-based prospective cohort study of time to pregnancy. Recruitment began in 2007 when an advertisement was placed on a Danish health-related website (www.netdoktor.dk) and a coordinated media strategy was launched (Mikkelsen et al., 2009; Rothman et al., 2009; Huybrechts et al., 2010). Enrollment and primary data collection were conducted via a self-administered questionnaire on the study website (www.snart-gravid.dk). Contact with participants was maintained through the study website and e-mail. Participants completed a consent form and an online screening questionnaire to verify eligibility for enrollment. Eligible women were aged 18 – 40 years, residents of Denmark, in a stable relationship with a male partner, not using fertility treatment, and trying to become pregnant. Participants were required to provide a valid e-mail address and their Civil Personal Registration (CPR) number, a unique 10-digit personal identification number
assigned to each Danish resident (Schmidt et al., 2014). After 38 months of recruitment, 5921 women had enrolled in the study. The study was approved by the Danish Data Protection Board and the Institutional Review Board of the Boston University Medical Campus. The baseline questionnaire collected information on demographics, lifestyle and behavioral factors, and reproductive and medical history. Participants were contacted every 2 months by e-mail with reminders to fill out a follow-up questionnaire. Follow-up questionnaires assessed changes in exposures and pregnancy status, including occurrence of any clinically recognized pregnancy losses. Follow-up continued until conception or for a maximum of 12 months. Women who were pregnant at the time of a followup questionnaire were asked to complete an early pregnancy questionnaire during their first trimester to assess any changes in exposures since conception as well as pregnancy symptoms. To obtain information on pregnancy outcomes among women in the cohort, we linked each woman’s CPR number to the Danish National Registry of Patients (DNRP) and the Danish Medical Birth Registry (DMBR). The DNRP provides information on hospitalizations and outpatient encounters (including SAB and therapeutic abortion) and the DMBR provides information on all live births and stillbirths after 22 gestational weeks (GW) (Kristensen et al., 1996; Lohse et al., 2010). International Classification of Disease (ICD), 10th edition, codes (DO03 for SAB and DO04 for therapeutic abortion) in the DNRP were used to identify pregnancy outcomes occurring among Snart-Gravid cohort members after the baseline enrollment date. A recent validation study comparing DNRP data on SABs with data from individual medical records found that the registry information had a positive predictive value of 98.7% (Lohse et al., 2010).
Assessment of SAB Women who experienced a pregnancy loss after enrollment were asked to report the date of the loss and gestational weeks at time of loss (time since the last menstrual period (LMP)) on a study questionnaire. The DNRP also provided information on SABs treated in hospital up to 22 gestational weeks and any therapeutic abortions, the dates of these events, and the gestational age at which the pregnancy ended. For pregnancy losses recorded in both the registry and on a questionnaire, we used data from the DNRP (based on either early ultrasound fetometry or LMP) to ascertain gestational week of pregnancy loss. In Denmark, the first pregnancy-related ultrasound is performed after about 12 weeks of gestation; gestational ages at SAB after this time are likely based on ultrasound. For SABs reported only on a Snart-Gravid follow-up questionnaire, gestational age was calculated as the number of weeks from the date of LMP to the date of pregnancy loss, rounded to the nearest whole week. For losses identified in both sources, the mean gestational age reported on Snart-Gravid questionnaires was slightly lower than that recorded in the registry (6.8 (SD: 1.8) weeks versus 7.2 weeks (SD: 1.7)) and did not vary appreciably by caffeine exposure. For 175 women, a pregnancy reported on a Snart-Gravid follow-up questionnaire had no corresponding data recorded in the hospital or birth registries. In these cases, we assumed that an early SAB had occurred. We used multiple imputation to impute a gestational age ≤12 weeks for each of these presumed SABs, under the assumption that later SABs would have been captured by the hospital registry. Sensitivity analyses excluding these pregnancies produced similar results (not shown).
Assessment of caffeine and caffeinated beverage consumption Servings per week of caffeinated coffee (250 ml mug), decaffeinated coffee (250 ml mug), herbal/green tea (250 ml mug), black tea (250 ml mug), regular cola (500 ml bottle) and diet cola (500 ml bottle) were reported on the baseline, follow-up and early pregnancy questionnaires. We used the following formula, based on milligrams (mg) of caffeine per serving of
Downloaded from http://humrep.oxfordjournals.org/ at University of California, San Francisco on April 17, 2015
et al., 2009; Schliep et al., 2012) have shown that caffeine consumption alters endogenous hormone levels. In some studies, caffeine has been inversely related to levels of estradiol and progesterone during the luteal phase of the menstrual cycle (Lawson et al., 2002; Kotsopoulos et al., 2009; Schliep et al., 2012) and positively related to sex hormone binding globulin (SHBG) (Kotsopoulos et al., 2009). Thus, hormonal changes related to caffeine consumption could plausibly affect the risk of spontaneous abortion (SAB). Published results of studies examining the association between caffeine and SAB risk have been conflicting (Fenster et al., 1991, 1997; Infante-Rivard et al., 1993; Mills et al., 1993; Dominguez-Rojas et al., 1994; Dlugosz et al., 1996; Cnattingius et al., 2000; Bech et al., 2005; Buss et al., 2006; Matijasevich et al., 2006; Savitz et al., 2008; Weng et al., 2008; Greenwood et al., 2010; Pollack et al., 2010), likely due to differences in study design and populations, outcome ascertainment, and exposure classification. Most previous studies of caffeine and SAB risk have focused on total caffeine intake from caffeinated beverage consumption during pregnancy. Some studies have also included chocolate (Wen et al., 2001) and caffeine from medications (Fenster et al., 1991; Cnattingius et al., 2000) while others have focused primarily on coffee (Armstrong et al., 1992; Dominguez-Rojas et al., 1994; Bech et al., 2005). In Danish prospective studies, the definition of high caffeine consumption has varied substantially, up to as much as ≥8 servings/day of coffee (Bech et al., 2005) and .900 mg/day of caffeine (Tolstrup et al., 2003). However, several other studies have used a definition of ≥300 mg/day (Mills et al., 1993; Dlugosz et al., 1996; Fenster et al., 1997; Wen et al., 2001; Greenwood et al., 2010). Most prospective studies (Dlugosz et al., 1996; Fenster et al., 1997; Bech et al., 2005; Savitz et al., 2008; Weng et al., 2008; Greenwood et al., 2010) have enrolled participants during early pregnancy and asked about caffeine consumption prior to the interview. This could lead to potential exposure misclassification, as well as differential left truncation bias. Prospective studies that have enrolled participants before pregnancy (Mills et al., 1993; Wen et al., 2001; Pollack et al., 2010) have been limited by small study sizes (431, 113 and 575, respectively). The objective of this study was to examine SAB risk in relation to consumption of caffeine and caffeinated beverages during preconception and early pregnancy among women enrolled in a large prospective cohort study in Denmark.
1248
Assessment of confounders Data on maternal age, parity, smoking status, prior SAB, alcohol consumption, physical activity, height and weight, and vocational training/education were self-reported on the baseline questionnaire. We estimated total metabolic equivalents (METs) per week by summing the METs from moderate physical activity (hours per week multiplied by 3.5 METs) and vigorous exercise (hours per week multiplied by 7.0 METs) (Jacobs et al., 1993). Body mass index (BMI) was calculated as kg/m2 and categorized as follows: ,20, 20 – 24, 25 – 29, ≥30. Study participants updated their smoking status and alcohol consumption on all subsequent follow-up questionnaires, including the early pregnancy questionnaire. We analyzed the covariate information from the questionnaire immediately before the pregnancy loss. The categories used for each confounder examined were as follows: age (18– 24, 25 – 29, 30 – 34, ≥35 years), smoking status (non-smoker, ,10 cigarettes/day, ≥10 cigarettes/day), alcohol consumption (0, 1, 2, 3 –6, 7 or more drinks/week), prior SAB (yes/no), physical activity (,10, 10 – 19, 20 – 39, ≥40 METs/week), vocational training/education (none, semiskilled/basic training, ≤4 years, .4 years), parity (nulliparous, 1 birth, .1 birth), BMI (,20, 20 – 24, 25 – 29, ≥30 kg/m2).
Study population The present analysis focuses on women with a clinically recognized pregnancy conceived after enrollment in Snart-Gravid. We excluded 126 who did not live in Denmark after enrollment, 10 women who did not provide a valid CPR, and 653 (11%) women who did not conceive after enrollment in the cohort (indicated by absence of any pregnancy reported on the followup questionnaires or recorded in the registries).
Data analysis We assessed the relationships between consumption of total caffeine and individual caffeinated beverages and SAB risk separately during preconception and early pregnancy, using a time-to-event analysis. Time to SAB was measured in gestational weeks. We categorized caffeine consumption as ,100, 100– 199, 200– 299 and ≥300 mg per day based on a previous manuscript from the Snart-Gravid study (Hatch et al., 2012) and similar categories examined in other caffeine and SAB studies (Dlugosz et al., 1996; Fenster et al., 1997; Wen et al., 2001; Greenwood et al., 2010). The categories of daily caffeinated beverage consumption were 0, 1, 2 and ≥3 servings for coffee and 0, 1 and ≥2 servings for cola, black tea and herbal/green tea. We examined the shape and magnitude of the relation between daily caffeine
consumption and SAB risk using restricted cubic splines (Durrleman and Simon, 1989). We used Cox proportional hazards regression models, with gestational weeks as the time scale, to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for caffeine and caffeinated beverage consumption associated with SAB. We assumed that there was a true but unknown ordering for tied event times and used the ‘exact’ option in SAS PROC PHREG (Therneau and Grambsch, 2000), which takes into account all possible orderings of event times. The HR is approximately equal to the average per-week risk of SAB for the exposed category divided by the corresponding risk for the reference category. Therapeutic abortions were censored at the week of pregnancy termination and pregnancies lasting more than 22 weeks were censored at 22 weeks. We identified potential confounders from variables associated both with SAB and caffeine consumption in our data. We also considered covariates meeting the criteria for confounding based on a review of the literature and the assessment of causal graphs (Rothman et al., 2008). Covariates included in both the preconception and early pregnancy models were maternal age, cigarette smoking, previous SAB, parity, vocational training/education, and physical activity. The preconception model also was adjusted for alcohol consumption. Models for the individual beverages (coffee, cola, herbal/green, and black tea) were mutually adjusted for each other (Willett, 1998). In secondary analyses, we stratified the data by waiting time to pregnancy (TTP), smoking status and BMI. Because it has been reported that women with viable pregnancies mayexperience more severe nausea symptoms and aversion to caffeinated beverages, we stratified our analyses on reported nausea during early pregnancy (‘Have you experienced nausea or vomiting with this pregnancy?’). In addition, we examined the relationship between caffeine consumption and SAB as a dichotomous outcome using log-binomial models. The etiology of pregnancy loss likely differs for early and late losses (Savitz et al., 2002; United States. Congress. Office of Technology Assessment., 1985), especially by karyotype, so we also examined the relationship between preconception caffeine consumption and timing of losses (loss during ,8 weeks of gestation and loss during ≥8 weeks of gestation). The choice of 8 weeks as a cut point for evaluating the timing of pregnancy loss was based on karyotype data showing a higher proportion of chromosomal abnormalities prior to 8 weeks (Klein and Stein, 1987). For this analysis, pregnant women were at risk for having an early loss, but only women who were still carrying a fetus by the end of 7 weeks were at risk for having a late loss (≥8 weeks). Those who were part of the first risk calculation but not part of the second are those women who experienced an SAB or a therapeutic abortion before 8 weeks of gestation. The risk period for the ‘late loss’ analysis began at 8 gestational weeks. We used multiple imputation methods to impute missing covariates, exposures, and outcome information (Zhou et al., 2001). Missing covariate data ranged from 0% for maternal age, time to pregnancy, and smoking status to 44% for number of glasses of dessert wine during early pregnancy. We imputed individual early pregnancy beverage frequencies for 2016 participants (39%). Missing data for preconception beverages ranged from 2% for coffee to 5% for decaf coffee. We used PROC MI to create 5 imputed datasets based on 46 variables in the imputation model. We combined coefficients and standard errors across the imputed datasets using PROC MIANALYZE. See Supplementary Table SI for details on percent missingness for the variables included in the imputation model. Departures from the proportional hazards assumption were assessed with log survivor plots. SAS statistical software (version 9.3, SAS Institute) was used for all analyses.
Results Among the 5132 women who conceived after enrolling in Snart-Gravid, 732 (14.3%) were identified as having an SAB. Overall, 25% of SABs were
Downloaded from http://humrep.oxfordjournals.org/ at University of California, San Francisco on April 17, 2015
each beverage as estimated in previous laboratory measurements (Caffeine content of food & drugs, 2014): total caffeine ¼ caffeinated coffee x (141 mg) + decaffeinated coffee x (5 mg) + black tea x (56 mg) + regular cola x (51 mg) + diet cola x (66 mg). Because the questionnaire did not distinguish between consumption of herbal tea and green tea, we did not include this item in our caffeine formula. Daily servings of each beverage also were considered individually in our analyses. We calculated a conception date for each participant based on the date of LMP, date of SAB and gestational age at SAB. For the analysis of preconception caffeine consumption, we analyzed exposure information that preceded the conception date. For those women who reported a pregnancy loss on a follow-up questionnaire, we relied on the most recent questionnaire before the reported loss for all exposure and covariate information. We analyzed data from the most recent questionnaire because pregnancy planners may change their caffeine consumption while attempting pregnancy (Lum et al., 2011). Early pregnancy caffeine consumption was based on self-reported information from before the loss for the 61% of women who provided information on beverage consumption in their early pregnancy questionnaire and was imputed for all other women.
Hahn et al.
1249
Caffeine consumption and spontaneous abortion
recorded only in the DNRP, 36% of SABs were reported only on a followup questionnaire, 15% of SABs were documented in both sources and 24% were pregnancies reported on a follow-up questionnaire but with no outcome information (live birth, SAB, therapeutic abortion) in either the DNRP or the DMBR. Baseline characteristics of study participants are presented in Table I. Preconception and early pregnancy caffeine consumption were associated with parity, higher education and cigarette smoking; preconception caffeine consumption was also positively associated with alcohol consumption. Most caffeine intake was from coffee; the Pearson correlation coefficient between servings of coffee consumed per day and total preconception caffeine consumption was 0.96. The correlation between number of
servings of coffee consumed per day during early pregnancy and total caffeine consumed during early pregnancy was 0.92, and the correlation between preconception caffeine consumption and caffeine consumption during early pregnancy was 0.48. After adjustment for all covariates, the HRs for preconception caffeine consumption of 100–299, 200–299 and ≥300 mg/day compared with ,100 mg/day were 1.00 (95% CI: 0.81, 1.23), 1.19 (95% CI: 0.96, 1.49) and 1.09 (95% CI: 0.89, 1.33), respectively (Table II). Figure 1 displays the relation between preconception caffeine consumption and risk of SAB by gestational week using a restricted cubic spline. The figure indicates little association between SAB risk and caffeine consumption at levels above 200 mg/day, consistent with the categorical results. Drinking ≥3 servings of coffee per day versus not drinking coffee was associated with a
Preconception caffeine consumption (mg/day)
................................................................
ORIGINAL ARTICLE Reproductive epidemiology
Caffeine and caffeinated beverage consumption and risk of spontaneous abortion 1
Department of Epidemiology, Boston University School of Public Health, Boston, MA 02118, USA 2Slone Epidemiology Center, Boston University, Boston, MA 02215, USA 3RTI Health Solutions, Research Triangle Park, NC 12194 USA 4Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark *Correspondence address. E-mail: [email protected]
Submitted on December 9, 2014; resubmitted on February 18, 2015; accepted on February 26, 2015
study question: Is caffeine and caffeinated beverage consumption associated with the risk of spontaneous abortion (SAB)? summary answer: While preconceptional caffeine consumption was not materially associated with an increased risk of SAB, consumption during early pregnancy was associated with a small increased risk of SAB, although the relation was not linear.
what is known already: Caffeine has been hypothesized as a risk factor for SAB since the 1980s; however, results from previous studies have been conflicting.
study design, size, duration: This prospective cohort study included 5132 Danish women planning pregnancy and enrolled from 2007 to 2010.
participants/materials, setting, methods: Participants were women who conceived after entry into the Snart-Gravid cohort and who were aged 18 –40, in a stable relationship with a male partner, and did not use fertility treatments to conceive. Women reported their daily caffeine and caffeinated beverage consumption on questionnaires before conception and during early pregnancy. All exposure measurements were prospective with respect to outcome ascertainment. We estimated hazard ratios (HRs) of SAB for categories of caffeine consumption in milligrams (mg) per day and the corresponding 95% confidence intervals (CIs) using Cox proportional hazards regression models with gestational weeks as the time scale. main results and the role of chance: There were 732 women (14.3%) who were identified as having a SAB. In the preconceptional period, caffeine consumption was not materially associated with SAB risk (HR comparing ≥300 with ,100 mg/day: 1.09; 95% CI: 0.89, 1.33). In early pregnancy, the HRs for 100 –199, 200 –299 and ≥300 mg/day of caffeine consumption were 1.62 (95% CI: 1.19, 2.22), 1.48 (95% CI: 1.03, 2.13) and 1.23 (95% CI: 0.61, 2.46), respectively, compared with that for ,100 mg/day. limitations, reasons for caution: The observed results may be affected by non-differential exposure misclassification, reverse causation and residual confounding.
wider implications of the findings: This is the largest study to date of prospectively measured, preconception caffeine consumption and risk of SAB. We were able to reduce the likelihood of differential left truncation bias and recall bias present in other analyses.
study funding/competing interest(s): Snart-Gravid was funded by the NICHD (R21-050264). Dr. Hahn’s work was funded in part by the BU Reproductive, Perinatal, and Pediatric Epidemiology Training Grant NIH #T32HD052458. There are no competing interests. Key words: caffeine / coffee / spontaneous abortion / cohort study
Introduction Caffeine crosses the placenta (Dlugosz and Bracken, 1992) and has a prolonged metabolism in pregnant women (15.08 hour half-life) compared
with non-pregnant women (4.71 hour half-life). Fetuses eliminate caffeine very slowly, suggesting that maternal caffeine ingestion could increase fetal caffeine levels exponentially (Brazier et al., 1983). Further, some studies (Lucero et al., 2001; Lawson et al., 2002; Kotsopoulos
& The Author 2015. Published by Oxford University Press on behalf of the European Society of Human Reproduction and Embryology. All rights reserved. For Permissions, please email: [email protected]
Downloaded from http://humrep.oxfordjournals.org/ at University of California, San Francisco on April 17, 2015
K.A. Hahn 1,*, L.A. Wise 1,2, K.J. Rothman 1,3, E.M. Mikkelsen4, S.B. Brogly 1, H.T. Sørensen 1,4, A.H. Riis 4, and E.E. Hatch 1
1247
Caffeine consumption and spontaneous abortion
Materials and Methods Data collection The Snart-Gravid study is an Internet-based prospective cohort study of time to pregnancy. Recruitment began in 2007 when an advertisement was placed on a Danish health-related website (www.netdoktor.dk) and a coordinated media strategy was launched (Mikkelsen et al., 2009; Rothman et al., 2009; Huybrechts et al., 2010). Enrollment and primary data collection were conducted via a self-administered questionnaire on the study website (www.snart-gravid.dk). Contact with participants was maintained through the study website and e-mail. Participants completed a consent form and an online screening questionnaire to verify eligibility for enrollment. Eligible women were aged 18 – 40 years, residents of Denmark, in a stable relationship with a male partner, not using fertility treatment, and trying to become pregnant. Participants were required to provide a valid e-mail address and their Civil Personal Registration (CPR) number, a unique 10-digit personal identification number
assigned to each Danish resident (Schmidt et al., 2014). After 38 months of recruitment, 5921 women had enrolled in the study. The study was approved by the Danish Data Protection Board and the Institutional Review Board of the Boston University Medical Campus. The baseline questionnaire collected information on demographics, lifestyle and behavioral factors, and reproductive and medical history. Participants were contacted every 2 months by e-mail with reminders to fill out a follow-up questionnaire. Follow-up questionnaires assessed changes in exposures and pregnancy status, including occurrence of any clinically recognized pregnancy losses. Follow-up continued until conception or for a maximum of 12 months. Women who were pregnant at the time of a followup questionnaire were asked to complete an early pregnancy questionnaire during their first trimester to assess any changes in exposures since conception as well as pregnancy symptoms. To obtain information on pregnancy outcomes among women in the cohort, we linked each woman’s CPR number to the Danish National Registry of Patients (DNRP) and the Danish Medical Birth Registry (DMBR). The DNRP provides information on hospitalizations and outpatient encounters (including SAB and therapeutic abortion) and the DMBR provides information on all live births and stillbirths after 22 gestational weeks (GW) (Kristensen et al., 1996; Lohse et al., 2010). International Classification of Disease (ICD), 10th edition, codes (DO03 for SAB and DO04 for therapeutic abortion) in the DNRP were used to identify pregnancy outcomes occurring among Snart-Gravid cohort members after the baseline enrollment date. A recent validation study comparing DNRP data on SABs with data from individual medical records found that the registry information had a positive predictive value of 98.7% (Lohse et al., 2010).
Assessment of SAB Women who experienced a pregnancy loss after enrollment were asked to report the date of the loss and gestational weeks at time of loss (time since the last menstrual period (LMP)) on a study questionnaire. The DNRP also provided information on SABs treated in hospital up to 22 gestational weeks and any therapeutic abortions, the dates of these events, and the gestational age at which the pregnancy ended. For pregnancy losses recorded in both the registry and on a questionnaire, we used data from the DNRP (based on either early ultrasound fetometry or LMP) to ascertain gestational week of pregnancy loss. In Denmark, the first pregnancy-related ultrasound is performed after about 12 weeks of gestation; gestational ages at SAB after this time are likely based on ultrasound. For SABs reported only on a Snart-Gravid follow-up questionnaire, gestational age was calculated as the number of weeks from the date of LMP to the date of pregnancy loss, rounded to the nearest whole week. For losses identified in both sources, the mean gestational age reported on Snart-Gravid questionnaires was slightly lower than that recorded in the registry (6.8 (SD: 1.8) weeks versus 7.2 weeks (SD: 1.7)) and did not vary appreciably by caffeine exposure. For 175 women, a pregnancy reported on a Snart-Gravid follow-up questionnaire had no corresponding data recorded in the hospital or birth registries. In these cases, we assumed that an early SAB had occurred. We used multiple imputation to impute a gestational age ≤12 weeks for each of these presumed SABs, under the assumption that later SABs would have been captured by the hospital registry. Sensitivity analyses excluding these pregnancies produced similar results (not shown).
Assessment of caffeine and caffeinated beverage consumption Servings per week of caffeinated coffee (250 ml mug), decaffeinated coffee (250 ml mug), herbal/green tea (250 ml mug), black tea (250 ml mug), regular cola (500 ml bottle) and diet cola (500 ml bottle) were reported on the baseline, follow-up and early pregnancy questionnaires. We used the following formula, based on milligrams (mg) of caffeine per serving of
Downloaded from http://humrep.oxfordjournals.org/ at University of California, San Francisco on April 17, 2015
et al., 2009; Schliep et al., 2012) have shown that caffeine consumption alters endogenous hormone levels. In some studies, caffeine has been inversely related to levels of estradiol and progesterone during the luteal phase of the menstrual cycle (Lawson et al., 2002; Kotsopoulos et al., 2009; Schliep et al., 2012) and positively related to sex hormone binding globulin (SHBG) (Kotsopoulos et al., 2009). Thus, hormonal changes related to caffeine consumption could plausibly affect the risk of spontaneous abortion (SAB). Published results of studies examining the association between caffeine and SAB risk have been conflicting (Fenster et al., 1991, 1997; Infante-Rivard et al., 1993; Mills et al., 1993; Dominguez-Rojas et al., 1994; Dlugosz et al., 1996; Cnattingius et al., 2000; Bech et al., 2005; Buss et al., 2006; Matijasevich et al., 2006; Savitz et al., 2008; Weng et al., 2008; Greenwood et al., 2010; Pollack et al., 2010), likely due to differences in study design and populations, outcome ascertainment, and exposure classification. Most previous studies of caffeine and SAB risk have focused on total caffeine intake from caffeinated beverage consumption during pregnancy. Some studies have also included chocolate (Wen et al., 2001) and caffeine from medications (Fenster et al., 1991; Cnattingius et al., 2000) while others have focused primarily on coffee (Armstrong et al., 1992; Dominguez-Rojas et al., 1994; Bech et al., 2005). In Danish prospective studies, the definition of high caffeine consumption has varied substantially, up to as much as ≥8 servings/day of coffee (Bech et al., 2005) and .900 mg/day of caffeine (Tolstrup et al., 2003). However, several other studies have used a definition of ≥300 mg/day (Mills et al., 1993; Dlugosz et al., 1996; Fenster et al., 1997; Wen et al., 2001; Greenwood et al., 2010). Most prospective studies (Dlugosz et al., 1996; Fenster et al., 1997; Bech et al., 2005; Savitz et al., 2008; Weng et al., 2008; Greenwood et al., 2010) have enrolled participants during early pregnancy and asked about caffeine consumption prior to the interview. This could lead to potential exposure misclassification, as well as differential left truncation bias. Prospective studies that have enrolled participants before pregnancy (Mills et al., 1993; Wen et al., 2001; Pollack et al., 2010) have been limited by small study sizes (431, 113 and 575, respectively). The objective of this study was to examine SAB risk in relation to consumption of caffeine and caffeinated beverages during preconception and early pregnancy among women enrolled in a large prospective cohort study in Denmark.
1248
Assessment of confounders Data on maternal age, parity, smoking status, prior SAB, alcohol consumption, physical activity, height and weight, and vocational training/education were self-reported on the baseline questionnaire. We estimated total metabolic equivalents (METs) per week by summing the METs from moderate physical activity (hours per week multiplied by 3.5 METs) and vigorous exercise (hours per week multiplied by 7.0 METs) (Jacobs et al., 1993). Body mass index (BMI) was calculated as kg/m2 and categorized as follows: ,20, 20 – 24, 25 – 29, ≥30. Study participants updated their smoking status and alcohol consumption on all subsequent follow-up questionnaires, including the early pregnancy questionnaire. We analyzed the covariate information from the questionnaire immediately before the pregnancy loss. The categories used for each confounder examined were as follows: age (18– 24, 25 – 29, 30 – 34, ≥35 years), smoking status (non-smoker, ,10 cigarettes/day, ≥10 cigarettes/day), alcohol consumption (0, 1, 2, 3 –6, 7 or more drinks/week), prior SAB (yes/no), physical activity (,10, 10 – 19, 20 – 39, ≥40 METs/week), vocational training/education (none, semiskilled/basic training, ≤4 years, .4 years), parity (nulliparous, 1 birth, .1 birth), BMI (,20, 20 – 24, 25 – 29, ≥30 kg/m2).
Study population The present analysis focuses on women with a clinically recognized pregnancy conceived after enrollment in Snart-Gravid. We excluded 126 who did not live in Denmark after enrollment, 10 women who did not provide a valid CPR, and 653 (11%) women who did not conceive after enrollment in the cohort (indicated by absence of any pregnancy reported on the followup questionnaires or recorded in the registries).
Data analysis We assessed the relationships between consumption of total caffeine and individual caffeinated beverages and SAB risk separately during preconception and early pregnancy, using a time-to-event analysis. Time to SAB was measured in gestational weeks. We categorized caffeine consumption as ,100, 100– 199, 200– 299 and ≥300 mg per day based on a previous manuscript from the Snart-Gravid study (Hatch et al., 2012) and similar categories examined in other caffeine and SAB studies (Dlugosz et al., 1996; Fenster et al., 1997; Wen et al., 2001; Greenwood et al., 2010). The categories of daily caffeinated beverage consumption were 0, 1, 2 and ≥3 servings for coffee and 0, 1 and ≥2 servings for cola, black tea and herbal/green tea. We examined the shape and magnitude of the relation between daily caffeine
consumption and SAB risk using restricted cubic splines (Durrleman and Simon, 1989). We used Cox proportional hazards regression models, with gestational weeks as the time scale, to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for caffeine and caffeinated beverage consumption associated with SAB. We assumed that there was a true but unknown ordering for tied event times and used the ‘exact’ option in SAS PROC PHREG (Therneau and Grambsch, 2000), which takes into account all possible orderings of event times. The HR is approximately equal to the average per-week risk of SAB for the exposed category divided by the corresponding risk for the reference category. Therapeutic abortions were censored at the week of pregnancy termination and pregnancies lasting more than 22 weeks were censored at 22 weeks. We identified potential confounders from variables associated both with SAB and caffeine consumption in our data. We also considered covariates meeting the criteria for confounding based on a review of the literature and the assessment of causal graphs (Rothman et al., 2008). Covariates included in both the preconception and early pregnancy models were maternal age, cigarette smoking, previous SAB, parity, vocational training/education, and physical activity. The preconception model also was adjusted for alcohol consumption. Models for the individual beverages (coffee, cola, herbal/green, and black tea) were mutually adjusted for each other (Willett, 1998). In secondary analyses, we stratified the data by waiting time to pregnancy (TTP), smoking status and BMI. Because it has been reported that women with viable pregnancies mayexperience more severe nausea symptoms and aversion to caffeinated beverages, we stratified our analyses on reported nausea during early pregnancy (‘Have you experienced nausea or vomiting with this pregnancy?’). In addition, we examined the relationship between caffeine consumption and SAB as a dichotomous outcome using log-binomial models. The etiology of pregnancy loss likely differs for early and late losses (Savitz et al., 2002; United States. Congress. Office of Technology Assessment., 1985), especially by karyotype, so we also examined the relationship between preconception caffeine consumption and timing of losses (loss during ,8 weeks of gestation and loss during ≥8 weeks of gestation). The choice of 8 weeks as a cut point for evaluating the timing of pregnancy loss was based on karyotype data showing a higher proportion of chromosomal abnormalities prior to 8 weeks (Klein and Stein, 1987). For this analysis, pregnant women were at risk for having an early loss, but only women who were still carrying a fetus by the end of 7 weeks were at risk for having a late loss (≥8 weeks). Those who were part of the first risk calculation but not part of the second are those women who experienced an SAB or a therapeutic abortion before 8 weeks of gestation. The risk period for the ‘late loss’ analysis began at 8 gestational weeks. We used multiple imputation methods to impute missing covariates, exposures, and outcome information (Zhou et al., 2001). Missing covariate data ranged from 0% for maternal age, time to pregnancy, and smoking status to 44% for number of glasses of dessert wine during early pregnancy. We imputed individual early pregnancy beverage frequencies for 2016 participants (39%). Missing data for preconception beverages ranged from 2% for coffee to 5% for decaf coffee. We used PROC MI to create 5 imputed datasets based on 46 variables in the imputation model. We combined coefficients and standard errors across the imputed datasets using PROC MIANALYZE. See Supplementary Table SI for details on percent missingness for the variables included in the imputation model. Departures from the proportional hazards assumption were assessed with log survivor plots. SAS statistical software (version 9.3, SAS Institute) was used for all analyses.
Results Among the 5132 women who conceived after enrolling in Snart-Gravid, 732 (14.3%) were identified as having an SAB. Overall, 25% of SABs were
Downloaded from http://humrep.oxfordjournals.org/ at University of California, San Francisco on April 17, 2015
each beverage as estimated in previous laboratory measurements (Caffeine content of food & drugs, 2014): total caffeine ¼ caffeinated coffee x (141 mg) + decaffeinated coffee x (5 mg) + black tea x (56 mg) + regular cola x (51 mg) + diet cola x (66 mg). Because the questionnaire did not distinguish between consumption of herbal tea and green tea, we did not include this item in our caffeine formula. Daily servings of each beverage also were considered individually in our analyses. We calculated a conception date for each participant based on the date of LMP, date of SAB and gestational age at SAB. For the analysis of preconception caffeine consumption, we analyzed exposure information that preceded the conception date. For those women who reported a pregnancy loss on a follow-up questionnaire, we relied on the most recent questionnaire before the reported loss for all exposure and covariate information. We analyzed data from the most recent questionnaire because pregnancy planners may change their caffeine consumption while attempting pregnancy (Lum et al., 2011). Early pregnancy caffeine consumption was based on self-reported information from before the loss for the 61% of women who provided information on beverage consumption in their early pregnancy questionnaire and was imputed for all other women.
Hahn et al.
1249
Caffeine consumption and spontaneous abortion
recorded only in the DNRP, 36% of SABs were reported only on a followup questionnaire, 15% of SABs were documented in both sources and 24% were pregnancies reported on a follow-up questionnaire but with no outcome information (live birth, SAB, therapeutic abortion) in either the DNRP or the DMBR. Baseline characteristics of study participants are presented in Table I. Preconception and early pregnancy caffeine consumption were associated with parity, higher education and cigarette smoking; preconception caffeine consumption was also positively associated with alcohol consumption. Most caffeine intake was from coffee; the Pearson correlation coefficient between servings of coffee consumed per day and total preconception caffeine consumption was 0.96. The correlation between number of
servings of coffee consumed per day during early pregnancy and total caffeine consumed during early pregnancy was 0.92, and the correlation between preconception caffeine consumption and caffeine consumption during early pregnancy was 0.48. After adjustment for all covariates, the HRs for preconception caffeine consumption of 100–299, 200–299 and ≥300 mg/day compared with ,100 mg/day were 1.00 (95% CI: 0.81, 1.23), 1.19 (95% CI: 0.96, 1.49) and 1.09 (95% CI: 0.89, 1.33), respectively (Table II). Figure 1 displays the relation between preconception caffeine consumption and risk of SAB by gestational week using a restricted cubic spline. The figure indicates little association between SAB risk and caffeine consumption at levels above 200 mg/day, consistent with the categorical results. Drinking ≥3 servings of coffee per day versus not drinking coffee was associated with a
Preconception caffeine consumption (mg/day)
................................................................
